1. Field of the Invention
The subject of the present invention is a method of hybridization of an inertial system by multi-hypothesis filtering as well as the associated hybrid inertial system. It applies notably within the framework of inertial systems, an aim of which is to determine the location and the navigation of a moving object, for example an aircraft. These systems comprise at least one inertial measurements unit whose outputs can have non-linear behaviours as well as a set of exterior sensors producing instantaneous measurements characterizing the location and/or the navigation of the said moving object. An application of such a system is, for example, the navigation of an aircraft.
2. Description of the Prior Art
An inertial measurement unit is traditionally composed of several inertial sensors which deliver one or more measurements describing the motion of the moving object. An inertial sensor can be a gyrometer which delivers a measurement of the angular speed of the moving object or an accelerometer which delivers a measurement of the acceleration of the moving object. These measurements may be marred by errors that will subsequently be called inertial defects and may, moreover, have a non-linear behaviour. For example, amplitude steps may be observed in the said measurements provided by the inertial sensor. These steps, or more generally, these non-linearities may be due to operating imperfections of the sensors, to high sensitivity to changes of temperature, to vibrations or else to the impact of abrupt changes of the motion profile of the moving object. One of the aims of the present invention is to implement a method for hybridizing an inertial system by instantaneous measurements of one or more components characteristic of the location and/or of the navigation of the moving object and provided by a set of external sensors.
Moreover, the invention proposes a solution which makes it possible to take into account the non-linear behaviour of the defects of the inertial sensors and is based on the use of a multi-hypothesis filter making it possible to identify these defects and ultimately to accurately correct the errors impacting the estimation of the location and of the navigation of the moving object. The multi-hypothesis filter according to the invention is notably based on the principle of making several hypotheses about the defects of the inertial sensors and then of determining which is the most likely.
The hybridization of inertial systems with other external sensors is traditionally carried out on the basis of extended Kalman filtering known by the abbreviation EKF (Extended Kalman Filter). This robust scheme is commonly used for numerous applications, notably for hybridizations performed with the aid of pressure sensors, magnetometers, satellite-based positioning systems, or else radioaltimeters. However, the proper operation of the extended Kalman filter relies on two hypotheses: the linearity of the propagation of its states and of its observations as well as the hypothesis that all noise affecting the system follows a Gaussian distribution. In particular, when the measurements provided by the inertial sensors do not have a linear behaviour, the extended Kalman filter is not suitable or remains largely sub-optimal.
International patent application WO00/73815 proposes a solution making it possible to alleviate the non-linearities of the defects of inertial sensors. This scheme is based on the use of neural networks and does not rely in any way on the principles of the extended Kalman filter. Moreover this application claims an inertial system using solely external sensors of GNSS type.